I wished to share how I exploit ChatGPT (and Bard in Google). I discover myself utilizing it day by day. I hope you discover a number of concepts right here which might be useful to you and share again a few of your finest prompts.
For me, generative AI is like hiring a fantastic knowledge scientist and a copywriter…plus having a pal who is aware of every part!
General, I exploit ChatGPT greater than Bard as I discover it offers higher solutions, summaries of my writing, and so forth. ChatGPT additionally has an ecosystem with 100+ plug-ins constructed round it. I’ve activated Noteable (creates a pocket book of Python code primarily based on statements in English of what I need to do), and the Wolfram plug-in. ChatGPT is aware of when to go management to the best plug-in primarily based on the immediate! For instance, I requested it for the legislation of cosines (gives measures of similarity between vectors) and it went to Wolfram Alpha for that.
Nonetheless, Bard can be very helpful particularly for information that require present information as it’s frequently refreshed (e.g. model promoting budgets). I additionally use it for Python code for Colab; since that may be a Google cloud product, I determine it would give me higher solutions there.
The most important methods Generative AI helps me:
Information science: ChatGPT writes code for me in R and Python (and even debugs error messages) which saves me hours. I inform it in English what I would like and it writes code together with what packages have to be put in…an actual magic wand. I may ask it for explanations of math instruments, properties of stat distributions, and even to seek out derivatives.
Copy editor: I are typically too technical in my first draft of any doc. I feed the doc to ChatGPT and ask it to assist me enhance. I ask it what technical phrases want less complicated descriptions…it calls them out and even gives these descriptions. It even created a glossary for me for a technical paper that was 85% there.
Sooner reality discovering than through search: Search offers hyperlinks however generative AI offers solutions! Nonetheless, generally you need to problem the information introduced and you’ll sometimes get totally different solutions.
Private pursuits: how one thing works, historic information, decoding medical prognosis phrases, baseball participant efficiency and sabermetrics definitions.
Let me take you thru some latest classes.
Media overlaps can contaminate promoting raise research with confounding elements. For instance, you suppose you’re evaluating on-line video publicity to no publicity, however you actually are evaluating on-line video + some portion of show (extremely correlated).
Cross-media attain (i.e. what % of shoppers see your advert SOMEWHERE) is the same problem however expressed relative to media planning and supply. Similar downside although…if any person noticed promoting a technique, what’s the likelihood they noticed it one other method too?
When you can estimate media overlaps, it will be of nice worth.
I got here up with an concept for tackling this downside by leveraging what a marketer might know…the attain of adverts on every platform individually and the correlations between publicity on the totally different platforms. Then I wished to create a mannequin that might simulate 10,000 digital IDs so I might analyze overlaps.
I requested ChatGPT about Copulas. These are features that comprise the correlational construction of marginal variables that may have utterly totally different distributions. The Copula plus its related marginal distributions offers you the power to simulate digital IDs that ought to comprise the best correlation patterns. At the least, that’s what I wished to check.
Now, that is the place ChatGPT actually helped. I figured there is perhaps a program in R to do what I had in thoughts. I instructed ChatGPT what I wished, and it created the code together with what packages I wanted to put in in R. It labored like a attraction! Then I simulated 10,000 information from my new operate and analyzed the outcomes. ChatGPT jogged my memory of the code to export the recordsdata as csv recordsdata. In Excel I made up my mind I acquired again the best correlations and now I used to be in enterprise. My 10,000 digital IDs gave me extremely correct cross-media publicity patterns.
Since this was such a brand new concept, I really debated it with ChatGPT by means of a sequence of prompts and responses that stored constructing on one another. The forwards and backwards actually had the texture of a dialogue with a colleague however ultimately, I used to be satisfied that I had a legitimate methodology.
Explaining this superior method to non-technical colleagues was not really easy so once I wrote up the methodology, I requested ChatGPT to simplify my language and make it comprehensible. I requested it for the 5 important concepts. I requested it for a title. All of this labored very effectively.
Now all work and no play isn’t nice so I interspersed some baseball stuff. I requested Bard which Yankees have the best profession OPS+ (on base share + slugging adjusted for year-to-year results.) I acquired my reply! (I used to be shocked to see Aaron Choose proper after Mantle however forward of DiMaggio!) Then I used to be on the lookout for a brand new present to binge, so I requested Bard for streaming video exhibits that had scores on rotten tomatoes over 90. I acquired a fantastic listing! The subsequent day, I acquired recipe concepts with Bulgar from ChatGPT.
The one restrict to ChatGPT is your creativeness. One researcher instructed me they even used it to create a code construction for open ended responses.
Please share a few of your finest tales of what labored (or didn’t) with ChatGPT.
